Please examine the following code:
if (foo->bar == NULL);
foo->bar = strdup("Unknown");
I spent the last part of three hours hunting down that leak with Valgrind, feeling very silly when I discovered the bogus ';'.
I know that the above code is valid C, however I would love for GCC to be able to tell me if I'm using a conditional as a statement.
Is there a flag that I could pass which would help spot this type of mistake in the future? It seems to me that GCC would be able to know if a conditional is useless.
That is,
if (1 == 1);
code_that_is_always_reached_since_conditional_is_a_statement();
None of the lints take issue with this either. Valgrind is great to find these kinds of things .. but the leak was actually much later in the code than where the pointer was originally allocated.
To summarize, here are your options:
- -Wextra picks up on all kinds of things that -Wall does not, including empty or useless statements.
- -Wempty-body picks up on useless statements, which is enabled by -Wextra (but it can break older versions of GCC, but it works on 4.3.x)
Some people might find -Wextra annoying. You may have a comparison between types of different signedness, but you know the comparison happens only when they are the same.
That is,
int ret;
unsigned int i;
ret = foo(bar); /* foo() is known to return a signed errno on failure */
if (ret < 0)
return 1;
/* Enter GCC complaining that ret differs in signedness
* (but you know it doesn't and get the urge to cast it) */
for (i = 0; i < ret; i ++)
...